In early 2024, a passenger named Jake Moffatt asked Air Canada’s website chatbot about bereavement fares. The bot told him he could book a regular ticket and apply for a discount retroactively. He did. Air Canada refused — pointing to a different policy page on their own site.
The British Columbia Civil Resolution Tribunal disagreed. It made Air Canada pay. The tribunal’s reasoning, in one sentence: the company is responsible for everything on its website, including what its chatbot says.
That ruling is two years old now, and the precedent has only widened. In the months since, U.S. plaintiffs’ counsel have brought parallel claims under state UDAP statutes against half a dozen companies whose AI agents made promises the company then declined to honour. None of those cases involve novel legal theory. They all reduce to the same principle: your AI is your speech.
What this means for your insurance programme
If your product, your sales funnel or your customer-service stack includes any generative AI component — and at this point nearly every firm we underwrite has one — the question is no longer whether an AI hallucination will create liability. It’s which policy responds when it does.
The honest answer is that most off-the-shelf programmes don’t respond cleanly.
- Standard cyber forms were written for breach response and ransomware. Some now include affirmative AI endorsements; many still treat AI-generated output as out of scope or sub-limit it to nuisance amounts.
- Tech E&O covers errors in the delivery of professional services. Whether an AI hallucination counts as a “professional service error” or a “product defect” is genuinely contested form-by-form.
- General liability picks up bodily injury and property damage. A bad chatbot quote that costs a customer money is neither.
- Media liability picks up defamation and IP infringement. That’s the closest cousin to chatbot fabrication, and the cleanest match for a hallucinated factual claim — but most operators don’t carry it standalone.
The market answer in 2026 is a dedicated AI liability endorsement, written onto cyber or tech E&O, that affirmatively covers (a) erroneous or fabricated output, (b) discriminatory or biased decisioning, and (c) IP infringement in training data or output. Several specialty carriers now offer this. Many don’t yet. The differences between forms are larger than the price differences.
The practical move for the next 90 days
Three things every firm with an AI surface should be doing right now:
- Audit where AI touches a customer decision. If a chatbot, recommendation engine or pricing algorithm is making a representation a reasonable person would rely on, that’s where the liability lives. Inventory it.
- Read your current cyber and Tech E&O forms. Specifically look for: AI exclusion (often added quietly at renewal), affirmative AI endorsement, and sub-limits on “media liability” or “personal & advertising injury.”
- Get the right endorsement on at renewal. If the underwriter is reluctant, that’s information — they’re probably right that the exposure is real, and a different carrier may price it differently.
The Air Canada case mattered because it was the first jurisdiction to say out loud that the company owns its AI’s outputs. The legal theory was unremarkable. What’s new is that the AI now talks to enough customers that one mistake — at scale, in writing — creates the same exposure that used to take a thousand customer service calls to generate.
If your firm ships AI, the insurance market in 2026 is treating it as a new line of risk. You should too.
